Assistant Professor in the Computer Science Department
at the University of New Hampshire
“This is how you do it”: Learning Concepts from Human Demonstrations
Learning from demonstration (LfD) is a rapidly emerging robot learning paradigm that promises the lay users with the ability to train a robot new tasks/skill without any knowledge of robotics or programming. There are two branches of LfD that have made tremendous progress in the past decade: learning low level motion primitives and learning high-level abstract reasoning. Low level LfD involves encoding of motion data to ensure robustness against perturbations such as change in start/goal position, unplanned obstacles, etc. High level LfD, on the other hand, involves ‘understanding the thought process’ of a demonstrator to develop a consistent representation of the task. In this talk, I will discuss the ongoing research at the Assistive Robotics Lab at UNH to bring both branches of LfD in the field. As for high-level LfD, I will present a deep Q-network architecture that we are developing to learn to deliver ABA-based autism intervention based on domain-experts’ demonstrations. As for low-level LfD, I will present a novel dynamic system based approach of trajectory learning that can be used to design home neuro-rehabilitation system for people with chronic moto disabilities.
Momotaz Begum (http://www.cs.unh.edu/~mbegum/) is an Assistant Professor in the Computer Science Department at the University of New Hampshire (UNH). Her research is on perception, learning, and human-robot interaction algorithms for intelligent assistive robots. Momotaz directs the Assistive Robotics Lab at UNH. Her research is sponsored through NSF and IEEE. Momotaz received her PhD on mobile robotics from the University of Waterloo and performed post-graduate research on human-robot interaction at the University of Toronto, Georgia Institute of Technology, and the University of Massachusetts Lowell.
Friday, April 27, 2018
2:00 p.m. - 3:00 p.m.
60 Gateway Park, GP 1002